Backup-bdg's picture
Upload 964 files
51ff9e5 verified
raw
history blame
31.8 kB
"""
This runtime runs commands locally using subprocess and performs file operations using Python's standard library.
It does not implement browser functionality.
"""
import asyncio
import os
import select
import shutil
import signal
import subprocess
import tempfile
import time
import zipfile
from pathlib import Path
from typing import Any, Callable
from binaryornot.check import is_binary
from openhands_aci.editor.editor import OHEditor
from openhands_aci.editor.exceptions import ToolError
from openhands_aci.editor.results import ToolResult
from openhands_aci.utils.diff import get_diff
from pydantic import SecretStr
from openhands.core.config import OpenHandsConfig
from openhands.core.config.mcp_config import MCPConfig, MCPStdioServerConfig
from openhands.core.exceptions import LLMMalformedActionError
from openhands.core.logger import openhands_logger as logger
from openhands.events import EventStream
from openhands.events.action import (
BrowseInteractiveAction,
BrowseURLAction,
CmdRunAction,
FileEditAction,
FileReadAction,
FileWriteAction,
IPythonRunCellAction,
)
from openhands.events.action.mcp import MCPAction
from openhands.events.event import FileEditSource, FileReadSource
from openhands.events.observation import (
CmdOutputObservation,
ErrorObservation,
FileEditObservation,
FileReadObservation,
FileWriteObservation,
Observation,
)
from openhands.integrations.provider import PROVIDER_TOKEN_TYPE
from openhands.runtime.base import Runtime
from openhands.runtime.plugins import PluginRequirement
class CLIRuntime(Runtime):
"""
A runtime implementation that runs commands locally using subprocess and performs
file operations using Python's standard library. It does not implement browser functionality.
Args:
config (OpenHandsConfig): The application configuration.
event_stream (EventStream): The event stream to subscribe to.
sid (str, optional): The session ID. Defaults to 'default'.
plugins (list[PluginRequirement] | None, optional): List of plugin requirements. Defaults to None.
env_vars (dict[str, str] | None, optional): Environment variables to set. Defaults to None.
status_callback (Callable | None, optional): Callback for status updates. Defaults to None.
attach_to_existing (bool, optional): Whether to attach to an existing session. Defaults to False.
headless_mode (bool, optional): Whether to run in headless mode. Defaults to False.
user_id (str | None, optional): User ID for authentication. Defaults to None.
git_provider_tokens (PROVIDER_TOKEN_TYPE | None, optional): Git provider tokens. Defaults to None.
"""
def __init__(
self,
config: OpenHandsConfig,
event_stream: EventStream,
sid: str = 'default',
plugins: list[PluginRequirement] | None = None,
env_vars: dict[str, str] | None = None,
status_callback: Callable[[str, str, str], None] | None = None,
attach_to_existing: bool = False,
headless_mode: bool = False,
user_id: str | None = None,
git_provider_tokens: PROVIDER_TOKEN_TYPE | None = None,
):
super().__init__(
config,
event_stream,
sid,
plugins,
env_vars,
status_callback,
attach_to_existing,
headless_mode,
user_id,
git_provider_tokens,
)
# Set up workspace
if self.config.workspace_base is not None:
logger.warning(
f'Workspace base path is set to {self.config.workspace_base}. '
'It will be used as the path for the agent to run in. '
'Be careful, the agent can EDIT files in this directory!'
)
self._workspace_path = self.config.workspace_base
else:
# Create a temporary directory for the workspace
self._workspace_path = tempfile.mkdtemp(
prefix=f'openhands_workspace_{sid}_'
)
logger.info(f'Created temporary workspace at {self._workspace_path}')
# Runtime tests rely on this being set correctly.
self.config.workspace_mount_path_in_sandbox = self._workspace_path
# Initialize runtime state
self._runtime_initialized = False
self.file_editor = OHEditor(workspace_root=self._workspace_path)
self._shell_stream_callback: Callable[[str], None] | None = None
logger.warning(
'Initializing CLIRuntime. WARNING: NO SANDBOX IS USED. '
'This runtime executes commands directly on the local system. '
'Use with caution in untrusted environments.'
)
async def connect(self) -> None:
"""Initialize the runtime connection."""
self.send_status_message('STATUS$STARTING_RUNTIME')
# Ensure workspace directory exists
os.makedirs(self._workspace_path, exist_ok=True)
# Change to the workspace directory
os.chdir(self._workspace_path)
if not self.attach_to_existing:
await asyncio.to_thread(self.setup_initial_env)
self._runtime_initialized = True
self.send_status_message('STATUS$CONTAINER_STARTED')
logger.info(f'CLIRuntime initialized with workspace at {self._workspace_path}')
def add_env_vars(self, env_vars: dict[str, Any]) -> None:
"""
Adds environment variables to the current runtime environment.
For CLIRuntime, this means updating os.environ for the current process,
so that subsequent commands inherit these variables.
This overrides the BaseRuntime behavior which tries to run shell commands
before it's initialized and modify .bashrc, which is not ideal for local CLI.
"""
if not env_vars:
return
# We log only keys to avoid leaking sensitive values like tokens into logs.
logger.info(
f'[CLIRuntime] Setting environment variables for this session: {list(env_vars.keys())}'
)
for key, value in env_vars.items():
if isinstance(value, SecretStr):
os.environ[key] = value.get_secret_value()
logger.warning(f'[CLIRuntime] Set os.environ["{key}"] (from SecretStr)')
else:
os.environ[key] = value
logger.debug(f'[CLIRuntime] Set os.environ["{key}"]')
# We don't use self.run() here because this method is called
# during initialization before self._runtime_initialized is True.
def _safe_terminate_process(self, process_obj, signal_to_send=signal.SIGTERM):
"""
Safely attempts to terminate/kill a process group or a single process.
Args:
process_obj: the subprocess.Popen object started with start_new_session=True
signal_to_send: the signal to send to the process group or process.
"""
pid = getattr(process_obj, 'pid', None)
if pid is None:
return
group_desc = (
'kill process group'
if signal_to_send == signal.SIGKILL
else 'terminate process group'
)
process_desc = (
'kill process' if signal_to_send == signal.SIGKILL else 'terminate process'
)
try:
# Try to terminate/kill the entire process group
logger.debug(f'[_safe_terminate_process] Original PID to act on: {pid}')
pgid_to_kill = os.getpgid(
pid
) # This might raise ProcessLookupError if pid is already gone
logger.debug(
f'[_safe_terminate_process] Attempting to {group_desc} for PID {pid} (PGID: {pgid_to_kill}) with {signal_to_send}.'
)
os.killpg(pgid_to_kill, signal_to_send)
logger.debug(
f'[_safe_terminate_process] Successfully sent signal {signal_to_send} to PGID {pgid_to_kill} (original PID: {pid}).'
)
except ProcessLookupError as e_pgid:
logger.warning(
f'[_safe_terminate_process] ProcessLookupError getting PGID for PID {pid} (it might have already exited): {e_pgid}. Falling back to direct kill/terminate.'
)
try:
if signal_to_send == signal.SIGKILL:
process_obj.kill()
else:
process_obj.terminate()
logger.debug(
f'[_safe_terminate_process] Fallback: Terminated {process_desc} (PID: {pid}).'
)
except Exception as e_fallback:
logger.error(
f'[_safe_terminate_process] Fallback: Error during {process_desc} (PID: {pid}): {e_fallback}'
)
except (AttributeError, OSError) as e_os:
logger.error(
f'[_safe_terminate_process] OSError/AttributeError during {group_desc} for PID {pid}: {e_os}. Falling back.'
)
# Fallback: try to terminate/kill the main process directly.
try:
if signal_to_send == signal.SIGKILL:
process_obj.kill()
else:
process_obj.terminate()
logger.debug(
f'[_safe_terminate_process] Fallback: Terminated {process_desc} (PID: {pid}).'
)
except Exception as e_fallback:
logger.error(
f'[_safe_terminate_process] Fallback: Error during {process_desc} (PID: {pid}): {e_fallback}'
)
except (KeyboardInterrupt, SystemExit):
raise
except Exception as e:
logger.error(f'Error: {e}')
def _execute_shell_command(
self, command: str, timeout: float
) -> CmdOutputObservation:
"""
Execute a shell command and stream its output to a callback function.
Args:
command: The shell command to execute
timeout: Timeout in seconds for the command
Returns:
CmdOutputObservation containing the complete output and exit code
"""
output_lines = []
timed_out = False
start_time = time.monotonic()
# Use shell=True to run complex bash commands
process = subprocess.Popen(
['bash', '-c', command],
stdout=subprocess.PIPE,
stderr=subprocess.STDOUT,
text=True,
bufsize=1, # Explicitly line-buffered for text mode
universal_newlines=True,
start_new_session=True,
)
logger.debug(
f'[_execute_shell_command] PID of bash -c: {process.pid} for command: "{command}"'
)
exit_code = None
try:
if process.stdout:
while process.poll() is None:
if (
timeout is not None
and (time.monotonic() - start_time) > timeout
):
logger.debug(
f'Command "{command}" timed out after {timeout:.1f} seconds. Terminating.'
)
# Attempt to terminate the process group (SIGTERM)
self._safe_terminate_process(
process, signal_to_send=signal.SIGTERM
)
timed_out = True
break
ready_to_read, _, _ = select.select([process.stdout], [], [], 0.1)
if ready_to_read:
line = process.stdout.readline()
if line:
logger.debug(f'LINE: {line}')
output_lines.append(line)
if self._shell_stream_callback:
self._shell_stream_callback(line)
# Attempt to read any remaining data from stdout
if process.stdout and not process.stdout.closed:
try:
while line:
line = process.stdout.readline()
if line:
logger.debug(f'LINE: {line}')
output_lines.append(line)
if self._shell_stream_callback:
self._shell_stream_callback(line)
except Exception as e:
logger.warning(
f'Error reading directly from stdout after loop for "{command}": {e}'
)
exit_code = process.returncode
# If timeout occurred, ensure exit_code reflects this for the observation.
if timed_out:
exit_code = -1
except Exception as e:
logger.error(
f'Outer exception in _execute_shell_command for "{command}": {e}'
)
if process and process.poll() is None:
self._safe_terminate_process(process, signal_to_send=signal.SIGKILL)
return CmdOutputObservation(
command=command,
content=''.join(output_lines) + f'\nError during execution: {e}',
exit_code=-1,
)
complete_output = ''.join(output_lines)
logger.debug(
f'[_execute_shell_command] Complete output for "{command}" (len: {len(complete_output)}): {complete_output!r}'
)
obs_metadata = {'working_dir': self._workspace_path}
if timed_out:
obs_metadata['suffix'] = (
f'[The command timed out after {timeout:.1f} seconds.]'
)
# exit_code = -1 # This is already set if timed_out is True
return CmdOutputObservation(
command=command,
content=complete_output,
exit_code=exit_code,
metadata=obs_metadata,
)
def run(self, action: CmdRunAction) -> Observation:
"""Run a command using subprocess."""
if not self._runtime_initialized:
return ErrorObservation(
f'Runtime not initialized for command: {action.command}'
)
if action.is_input:
logger.warning(
f"CLIRuntime received an action with `is_input=True` (command: '{action.command}'). "
'CLIRuntime currently does not support sending input or signals to active processes. '
'This action will be ignored and an error observation will be returned.'
)
return ErrorObservation(
content=f"CLIRuntime does not support interactive input from the agent (e.g., 'C-c'). The command '{action.command}' was not sent to any process.",
error_id='AGENT_ERROR$BAD_ACTION',
)
try:
effective_timeout = (
action.timeout
if action.timeout is not None
else self.config.sandbox.timeout
)
logger.debug(
f'Running command in CLIRuntime: "{action.command}" with effective timeout: {effective_timeout}s'
)
return self._execute_shell_command(
action.command, timeout=effective_timeout
)
except Exception as e:
logger.error(
f'Error in CLIRuntime.run for command "{action.command}": {str(e)}'
)
return ErrorObservation(
f'Error running command "{action.command}": {str(e)}'
)
def run_ipython(self, action: IPythonRunCellAction) -> Observation:
"""Run a Python code cell.
This functionality is not implemented in CLIRuntime.
Users should also disable the Jupyter plugin in AgentConfig.
"""
# This functionality is not implemented in CLIRuntime.
# If you need to run IPython/Jupyter cells, please consider using a different runtime
# or ensure the Jupyter plugin is disabled in your AgentConfig to avoid
# attempting to use this disabled feature.
logger.warning(
"run_ipython is called on CLIRuntime, but it's not implemented. "
'Please disable the Jupyter plugin in AgentConfig.'
)
return ErrorObservation(
'Executing IPython cells is not implemented in CLIRuntime. '
)
def _sanitize_filename(self, filename: str) -> str:
# if path is absolute, ensure it starts with _workspace_path
if filename == '/workspace':
actual_filename = self._workspace_path
elif filename.startswith('/workspace/'):
# Map /workspace/ to the actual workspace path
# Note: /workspace is widely used, so we map it to allow using it with CLIRuntime
actual_filename = os.path.join(
self._workspace_path, filename[len('/workspace/') :]
)
elif filename.startswith('/'):
if not filename.startswith(self._workspace_path):
raise LLMMalformedActionError(
f'Invalid path: {filename}. You can only work with files in {self._workspace_path}.'
)
actual_filename = filename
else:
actual_filename = os.path.join(self._workspace_path, filename.lstrip('/'))
# Resolve the path to handle any '..' or '.' components
resolved_path = os.path.realpath(actual_filename)
# Check if the resolved path is still within the workspace
if not resolved_path.startswith(self._workspace_path):
raise LLMMalformedActionError(
f'Invalid path traversal: {filename}. Path resolves outside the workspace. Resolved: {resolved_path}, Workspace: {self._workspace_path}'
)
return resolved_path
def read(self, action: FileReadAction) -> Observation:
"""Read a file using Python's standard library or OHEditor."""
if not self._runtime_initialized:
return ErrorObservation('Runtime not initialized')
file_path = self._sanitize_filename(action.path)
# Cannot read binary files
if os.path.exists(file_path) and is_binary(file_path):
return ErrorObservation('ERROR_BINARY_FILE')
# Use OHEditor for OH_ACI implementation source
if action.impl_source == FileReadSource.OH_ACI:
result_str, _ = self._execute_file_editor(
command='view',
path=file_path,
view_range=action.view_range,
)
return FileReadObservation(
content=result_str,
path=action.path,
impl_source=FileReadSource.OH_ACI,
)
try:
# Check if the file exists
if not os.path.exists(file_path):
return ErrorObservation(f'File not found: {action.path}')
# Check if it's a directory
if os.path.isdir(file_path):
return ErrorObservation(f'Cannot read directory: {action.path}')
# Read the file
with open(file_path, 'r', encoding='utf-8', errors='replace') as f:
content = f.read()
return FileReadObservation(content=content, path=action.path)
except Exception as e:
logger.error(f'Error reading file: {str(e)}')
return ErrorObservation(f'Error reading file {action.path}: {str(e)}')
def write(self, action: FileWriteAction) -> Observation:
"""Write to a file using Python's standard library."""
if not self._runtime_initialized:
return ErrorObservation('Runtime not initialized')
file_path = self._sanitize_filename(action.path)
try:
# Create parent directories if they don't exist
os.makedirs(os.path.dirname(file_path), exist_ok=True)
# Write to the file
with open(file_path, 'w', encoding='utf-8') as f:
f.write(action.content)
return FileWriteObservation(content='', path=action.path)
except Exception as e:
logger.error(f'Error writing to file: {str(e)}')
return ErrorObservation(f'Error writing to file {action.path}: {str(e)}')
def browse(self, action: BrowseURLAction) -> Observation:
"""Not implemented for CLI runtime."""
return ErrorObservation(
'Browser functionality is not implemented in CLIRuntime'
)
def browse_interactive(self, action: BrowseInteractiveAction) -> Observation:
"""Not implemented for CLI runtime."""
return ErrorObservation(
'Browser functionality is not implemented in CLIRuntime'
)
def _execute_file_editor(
self,
command: str,
path: str,
file_text: str | None = None,
view_range: list[int] | None = None,
old_str: str | None = None,
new_str: str | None = None,
insert_line: int | None = None,
enable_linting: bool = False,
) -> tuple[str, tuple[str | None, str | None]]:
"""Execute file editor command and handle exceptions.
Args:
command: Editor command to execute
path: File path
file_text: Optional file text content
view_range: Optional view range tuple (start, end)
old_str: Optional string to replace
new_str: Optional replacement string
insert_line: Optional line number for insertion
enable_linting: Whether to enable linting
Returns:
tuple: A tuple containing the output string and a tuple of old and new file content
"""
result: ToolResult | None = None
try:
result = self.file_editor(
command=command,
path=path,
file_text=file_text,
view_range=view_range,
old_str=old_str,
new_str=new_str,
insert_line=insert_line,
enable_linting=enable_linting,
)
except ToolError as e:
result = ToolResult(error=e.message)
if result.error:
return f'ERROR:\n{result.error}', (None, None)
if not result.output:
logger.warning(f'No output from file_editor for {path}')
return '', (None, None)
return result.output, (result.old_content, result.new_content)
def edit(self, action: FileEditAction) -> Observation:
"""Edit a file using the OHEditor."""
if not self._runtime_initialized:
return ErrorObservation('Runtime not initialized')
# Ensure the path is within the workspace
file_path = self._sanitize_filename(action.path)
# Check if it's a binary file
if os.path.exists(file_path) and is_binary(file_path):
return ErrorObservation('ERROR_BINARY_FILE')
assert action.impl_source == FileEditSource.OH_ACI
result_str, (old_content, new_content) = self._execute_file_editor(
command=action.command,
path=file_path,
file_text=action.file_text,
old_str=action.old_str,
new_str=action.new_str,
insert_line=action.insert_line,
enable_linting=False,
)
return FileEditObservation(
content=result_str,
path=action.path,
old_content=action.old_str,
new_content=action.new_str,
impl_source=FileEditSource.OH_ACI,
diff=get_diff(
old_contents=old_content or '',
new_contents=new_content or '',
filepath=action.path,
),
)
async def call_tool_mcp(self, action: MCPAction) -> Observation:
"""Not implemented for CLI runtime."""
return ErrorObservation('MCP functionality is not implemented in CLIRuntime')
@property
def workspace_root(self) -> Path:
"""Return the workspace root path."""
return Path(os.path.abspath(self._workspace_path))
def copy_to(self, host_src: str, sandbox_dest: str, recursive: bool = False):
"""Copy a file or directory from the host to the sandbox."""
if not self._runtime_initialized:
raise RuntimeError('Runtime not initialized')
if not os.path.exists(host_src): # Source must exist on host
raise FileNotFoundError(f"Source path '{host_src}' does not exist.")
dest = self._sanitize_filename(sandbox_dest)
try:
# Case 1: Source is a directory and recursive copy.
if os.path.isdir(host_src) and recursive:
# Target is dest / basename(host_src)
final_target_dir = os.path.join(dest, os.path.basename(host_src))
# If source and final target are the same, skip.
if os.path.realpath(host_src) == os.path.realpath(final_target_dir):
logger.debug(
'Skipping recursive copy: source and target are identical.'
)
pass
else:
# Ensure parent of final_target_dir exists.
os.makedirs(dest, exist_ok=True)
shutil.copytree(host_src, final_target_dir, dirs_exist_ok=True)
# Why: Copies dir host_src into dest. Merges if target exists.
# Case 2: Source is a file.
elif os.path.isfile(host_src):
final_target_file_path: str
# Scenario A: sandbox_dest is clearly a directory.
if os.path.isdir(dest) or (sandbox_dest.endswith(('/', os.sep))):
target_dir = dest
os.makedirs(target_dir, exist_ok=True)
final_target_file_path = os.path.join(
target_dir, os.path.basename(host_src)
)
# Why: Copies file into specified directory.
# Scenario B: sandbox_dest is likely a new directory (e.g., 'new_dir').
elif not os.path.exists(dest) and '.' not in os.path.basename(dest):
target_dir = dest
os.makedirs(target_dir, exist_ok=True)
final_target_file_path = os.path.join(
target_dir, os.path.basename(host_src)
)
# Why: Creates 'new_dir' and copies file into it.
# Scenario C: sandbox_dest is a full file path.
else:
final_target_file_path = dest
os.makedirs(os.path.dirname(final_target_file_path), exist_ok=True)
# Why: Copies file to a specific path, possibly renaming.
shutil.copy2(host_src, final_target_file_path)
else: # Source is not a valid file or directory.
raise FileNotFoundError(
f"Source path '{host_src}' is not a valid file or directory."
)
except FileNotFoundError as e:
logger.error(f'File not found during copy: {str(e)}')
raise
except shutil.SameFileError as e:
# We can be lenient here, just ignore this error.
logger.debug(
f'Skipping copy as source and destination are the same: {str(e)}'
)
pass
except Exception as e:
logger.error(f'Unexpected error copying file: {str(e)}')
raise RuntimeError(f'Unexpected error copying file: {str(e)}')
def list_files(self, path: str | None = None) -> list[str]:
"""List files in the sandbox."""
if not self._runtime_initialized:
raise RuntimeError('Runtime not initialized')
if path is None:
dir_path = self._workspace_path
else:
dir_path = self._sanitize_filename(path)
try:
if not os.path.exists(dir_path):
return []
if not os.path.isdir(dir_path):
return [dir_path]
# List files in the directory
return [os.path.join(dir_path, f) for f in os.listdir(dir_path)]
except Exception as e:
logger.error(f'Error listing files: {str(e)}')
return []
def copy_from(self, path: str) -> Path:
"""Zip all files in the sandbox and return a path in the local filesystem."""
if not self._runtime_initialized:
raise RuntimeError('Runtime not initialized')
source_path = self._sanitize_filename(path)
if not os.path.exists(source_path):
raise FileNotFoundError(f'Path not found: {path}')
# Create a temporary zip file
temp_zip = tempfile.NamedTemporaryFile(suffix='.zip', delete=False)
temp_zip.close()
try:
with zipfile.ZipFile(temp_zip.name, 'w', zipfile.ZIP_DEFLATED) as zipf:
if os.path.isdir(source_path):
# Add all files in the directory
for root, _, files in os.walk(source_path):
for file in files:
file_path = os.path.join(root, file)
arcname = os.path.relpath(file_path, source_path)
zipf.write(file_path, arcname)
else:
# Add a single file
zipf.write(source_path, os.path.basename(source_path))
return Path(temp_zip.name)
except Exception as e:
logger.error(f'Error creating zip file: {str(e)}')
raise RuntimeError(f'Error creating zip file: {str(e)}')
def close(self) -> None:
self._runtime_initialized = False
super().close()
@classmethod
async def delete(cls, conversation_id: str) -> None:
"""Delete any resources associated with a conversation."""
# Look for temporary directories that might be associated with this conversation
temp_dir = tempfile.gettempdir()
prefix = f'openhands_workspace_{conversation_id}_'
for item in os.listdir(temp_dir):
if item.startswith(prefix):
try:
path = os.path.join(temp_dir, item)
if os.path.isdir(path):
shutil.rmtree(path)
logger.info(f'Deleted workspace directory: {path}')
except Exception as e:
logger.error(f'Error deleting workspace directory: {str(e)}')
@property
def additional_agent_instructions(self) -> str:
return '\n\n'.join(
[
f'Your working directory is {self._workspace_path}. You can only read and write files in this directory.',
"You are working directly on the user's machine. In most cases, the working environment is already set up.",
]
)
def get_mcp_config(
self, extra_stdio_servers: list[MCPStdioServerConfig] | None = None
) -> MCPConfig:
# TODO: Load MCP config from a local file
return MCPConfig()
def subscribe_to_shell_stream(
self, callback: Callable[[str], None] | None = None
) -> bool:
"""
Subscribe to shell command output stream.
Args:
callback: A function that will be called with each line of output from shell commands.
If None, any existing subscription will be removed.
"""
self._shell_stream_callback = callback
return True